1

I am trying to make Mathematica recognize my OpenCL installation, which is beignet. It is experimental software, so I know that it probably won't work (even clinfo terminates with an error after displaying some info), yet I would like to know exactly why OpenCLQ[] returns False.

Lorenzo Pistone
  • 1,607
  • 11
  • 16
  • Does Trace or TracePrint help? – Sjoerd C. de Vries Jan 15 '14 at 12:41
  • 1
    I think it would probably help to look at the source code for the OpenCLLink package ($InstallationDirectory`SystemFiles`Links`OpenCLLink) and follow the logic through. I had trouble with OpenCL in the past and did so myself; there seemed to be quite a few strange assumptions/incorrect logic involved in it, at least in version 8. This may have been improved but I have the feeling it would still not be completely robust. Anyway, this should give you an idea of what is needed for OpenCLQ[] to return True. – Oleksandr R. Jan 15 '14 at 14:55
  • @SjoerdC.deVries it's a load of text, not really readable – Lorenzo Pistone Jan 15 '14 at 22:58
  • @OleksandrR. I can't find the logic in the .m files, or am I looking i nthe wrong place? I fear that the logic is in the binaries – Lorenzo Pistone Jan 15 '14 at 22:59
  • Sorry, yes, the layout of these packages seems to have changed (or I misremembered). The logic behind PlatformCount (on which I suppose you saw that OpenCLQ depends) is in $InstallationDirectory`SystemFiles`Links`GPUTools`WGLPrivate.m. – Oleksandr R. Jan 16 '14 at 04:16

0 Answers0